The Agent Skills Directory

[PROMPT_INJECTION]: The skill mandates that the agent copy the 'Original User Request' verbatim into the plan.md file (defined in SKILL.md and references/04_TEMPLATES.md). Because these planning documents are intended to be read and followed by other agents for multi-agent coordination (as detailed in references/06_MULTI_AGENT_COORDINATION.md), this creates a vector for indirect prompt injection. A malicious user could provide a request containing hidden instructions that would then be persisted in the planning system and executed by future agents during their scanning phase.\n
Ingestion points: SKILL.md (Original User Request section) and references/04_TEMPLATES.md (Root Plan Template).\n
Boundary markers: None. The instructions explicitly state to paste the text 'unchanged (no paraphrasing or summary)'.\n
Capability inventory: File system writes (docs/planning/), and shell command execution (ls, git).\n
Sanitization: Explicitly absent; the skill requires exact wording.\n- [COMMAND_EXECUTION]: The skill executes several shell commands to manage state and coordinate between agents, such as ls -dt docs/planning/phase-*, git status --porcelain, git diff --stat, and git log. These are used for conflict detection and repository management as part of the intended workflow.

phase-plan